Research map / Detection and segmentation / CNN features
Salient Object Ranking With Position-Preserved Attention
Hao Fang, Daoxin Zhang, Yi Zhang, Minghao Chen, Jiawei Li, Yao Hu, Deng Cai, Xiaofei He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train end-to-end instead of two-stage (ablation) | PPA SOR network · ASR dataset | SOR: 0.835 -> 0.841 | Improved |
| train end-to-end instead of two-stage (ablation) | PPA SOR network · ASR dataset | MAE (lower is better): 0.082 -> 0.081 | Improved |
| add position embedding stage to baseline SOR branch (ablation) | Baseline SOR branch · ASR dataset | MAE (lower is better): 0.104 -> 0.095 | Improved |
| add position embedding stage to baseline SOR branch (ablation) | Baseline SOR branch · ASR dataset | SOR: 0.830 -> 0.836 | Improved |
| add feature interaction (Transformer attention) stage to baseline (ablation) | Baseline SOR branch · ASR dataset | MAE (lower is better): 0.104 -> 0.088 | Improved |
| add feature interaction (Transformer attention) stage to baseline (ablation) | Baseline SOR branch · ASR dataset | SOR: 0.830 -> 0.839 | Improved |
| add both position embedding and attention to baseline (ablation) | Baseline SOR branch · ASR dataset | MAE (lower is better): 0.104 -> 0.081 | Improved |
| add attention on top of position embedding (ablation) | Baseline+pos · ASR dataset | SOR: 0.836 -> 0.841 | Improved |
| use PPA position embedding instead of concatenated center coordinates (ablation) | SOR branch · ASR dataset | MAE (lower is better): 0.092 -> 0.081 | Improved |
| add scale (w, h) to concatenated position information (ablation) | SOR branch with C_x,C_y · ASR dataset | MAE (lower is better): 0.092 -> 0.088 | Improved |
| use learnable look-up position embedding instead of PPA coordinate maps (ablation) | SOR branch · ASR dataset | SOR: 0.841 -> 0.821 | Hurt |
| add scale to learnable position embedding (ablation) | Learnable pos · ASR dataset | SOR: 0.821 -> 0.834 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- BASNet — BASNet: Boundary-Aware Salient Object Detection (CVPR 2019) · better on 2 of 3 shared results
- S4Net — S4Net: Single Stage Salient-Instance Segmentation (CVPR 2019) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Instance-Level Relative Saliency Ranking with Graph Reasoning (TPAMI 2021 · similarity 0.92)
- Relative Saliency and Ranking: Models, Metrics, Data and Benchmarks (TPAMI 2019 · similarity 0.92)
- Revisiting Salient Object Detection: Simultaneous Detection, Ranking, and Subitizing of Multiple Salient Objects (CVPR 2018 · similarity 0.92)
- SeqRank: Sequential Ranking of Salient Objects (AAAI 2024 · similarity 0.91)
- Language-Guided Salient Object Ranking (CVPR 2025 · similarity 0.90)
- Advancing Saliency Ranking with Human Fixations: Dataset Models and Benchmarks (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
